The present invention relates to the electrical, electronic and computer arts, and more specifically, to conversations generated by conversational bots.
Conversational bots are becoming ubiquitous in business applications, social media, and the like. Identifying problems in conversational bots quickly and consistently is pertinent to the development and general acceptance of such bots. This is especially true given the variety of types of conversations that can be encountered. Discordant conversations, for example, are relatively rare (around 8% of conversations), but the detection of such problematic conversations in bots often proves difficult. For example, a number of previous efforts have focused on the task of dialogue breakdown detection, where an utterance by the bot that leads to dialogue breakdown, and a situation where the conversation with a user cannot proceed, is detected. These efforts do not necessarily address the exploration, visualization, and improvement of bot conversations, or the analysis and integration of emotions in bot conversations.